Info: I do not own this machine anymore, therefore my ability to answer your questions is limited.
In 2017 I bought a refurbished HP z820 Workstation sporting 2 unreasonably powerful electric heaters in the form of Intel Xeon E5-2687W v2 and comfortable 128 GB ECC memory. Since I used the machine for offloading CPU-based raytracing jobs only, I was able to store it in a more remote part of my apartment where it did not really bother me.
Now, the jobs I did back in the day are all done, and I don’t use the machine that often anymore. A shame really. Or, at least, I used to not use it anymore. But since my daily driver – a very particular combination of a Ryzen 5 1600, G.Skill RAM and an RX 480 – doesn’t play too well together with Linux kernel 5.x (Manjaro), presenting me with totally cool random reboots, system freezes and the like, I needed to look for alternatives.
So, the idea was formed to use the z820 instead. If it just wasn’t for the noise level. Ohh, the noise this thing makes. You see, I understand that the 2x 2687W v2 with 150W TDP each, high current VRMs, 128 GB ECC DDR3 and an RX 480 need proper cooling. But running this beast of a machine in its current state and under my desk was just out of the question.
An avid PC builder and quiet-loving enthusiast myself, I looked for a solution.
A typical HP z820 is cooled by:
- 2x 92mm axial case fans behind the drive cage, in a pull-configuration
Nidec T92T12MS3A7 – 57 A03
HP ASSY PN 647113-001 REV A
12V DC, max. 0.35 A
each 4-Pin, presumably standard pinout PWM driven
- 2x 92mm axial case fans at the rear of the case in a push-configuration
HP ASSY 644315-001 REV 0A
12V DC, max. 0.6A (-BL3H)
bundled together via a proprietary 6-pin contraption, presumably PWM driven
- 2x 80mm axial fans in the front of the PSU, in a pull-configuration
12V DC, max 0.5A (-HE00)
each 4-pin, presumably non-standard pinout, PWM driven
and this monstrosity here:
This is HPs idea of a wholly integrated fan assembly (‘shroud’) that holds 6 fans in total and is designed to cool the CPU, RAM banks and VRMs altogether. It is made up of
- 2x 92mm axial fans, designed to cool the CPUs
Nidec T92T12MS3A7 – 57 A03
HP ASSY PN 647113-001 REV A
12V DC, max. 0.35 A
each 4-Pin, presumably standard pinout PWM driven
- 3x OEM radial fans, Delta BUB0712HF, designed to cool the RAM banks
HP ASSY PN 670051-001 Rev A
12V DC, max 0.86 A (-BE04)
- 1x 60mm axial fan, designed to cool the VRMs
HP P/N 670050-001 Rev. A
12V DC, max. 0.7 A
These six fans are each connected via presumably standard 4-pin PWM connectors to this breakout cable (visible in the top left corner of ‘Exhibit A’)
that is in turn connected to the mainboard via its proprietary connector. While I am not a big fan of proprietary solutions, this breakout cable does at least provide a central point to hook up all the fans and allows for connecting up to 6 fans. That is not too shabby.
As it is now, the two 92mm CPU fans blow the air in a 45°-ish diagonal fashion through the fins of the OEM AIO liquid cooler towards the mainboard and the RAM, creating unwanted air turbulences and therefore making for a not-so-optimal situation. The additional radial fans that blow towards the RAM banks add to the turbulences and prevent an effective directional air flow.
Now, I understand that HP engineers had a clear vision of compartmentalized cooling/air flow. Having the CPU/RAM/VRMs in their separate cooling zone is clever and should indeed be implemented by all OEMs and should be an industry standard by now. The only thing that irritates me is the choice of hardware here, but that is something that would not have bothered me back in the day, but definitely bothers me now.
Instead of 2x 92 mm CPU fans, today I would have chosen 2x 120mm fans in combination with a grotesquely overdimensioned tower cooler, as is standard today. Properly placed, these could make use of the intake air flow from the front and pass the air out through to the back. Increased size of the fans allows for the fans to rotate slower as the 92mm ones while still providing the same (or even better) air flow and lower noise. Cooling down VRMs and RAM banks could be achieved by repurposing the now unused 5.25″ drive bays into one big cooling duct. I think placing one (or two?) 120mm fan(s) that pulls in air at the front of the case could prove to be very effective.
One could argue that it would be simpler (and much cheaper) to just replace the current fans with aftermarket ones. Well, that would indeed spare me and you lots of work but being overly compatible is not a OEMs primary intention. Unfortunately, it would prove to be partially very difficult to source more quiet replacements for the fans: the radial fans are proprietary OEM solutions and therefore exist no third-party alternatives. And just replacing two 92mm fans just doesn’t cut it.
Luckily, this does not apply to the axial fans. Apart from the plug, which seems to be a MOLEX KK 254 series plug (or some compatible equivalent) these fans seem to be pretty standard regarding dimensions, voltage and pinout and should be replaceable with compatible equivalents.
One notable example are the PSUs fans – here the pinout appears to be reversed. This needs to be further examined:
google brute forcing investigating this phenomenon, I stumbled across an image titled “Delta DPS-1050DB A_Fan_Pinout.png” hosted on HPs Servers. Unfortunately I was not able to trace that image back to a discussion thread or something similar to get some context. So take this with a rock grain of salt:
After some googling it was reavealed that – if one may trust the diagram and the associated title – the diagram (DPS-1050DB A) depicts the older z800’s 1100W PSU fan pinouts. Since the manufacturer (Delta) is the same and since the z800 shares many similarities with the z820 anyway, this further strengthens the theory that for the z820s PSUs fans we might just get away by reversing the pin order. We will have to find out by trying it out, eh.
So, my assumption about the PSUs fan pinout must have been somewhat wrong. Trying with either pinout (PWM-TACH-12V-GND and GND-12V-TACH-PWM) yielded a non-booting machine and an error code: the power button lit up in red and the machine beeped 4 times with 1 seconds delays. The manual giveth us knowledge: ‘Power failure (powersupply is overloaded).’
Even though the fans I tried were 0.07A fans, it was too much for the 1125W PSU. This can obviously not the reason for the failure.
So, obviously the PSUs fans must be a special kind of fan. Or have a very unusual pinout. My best guess is that the fans themselves sport some kind of functionality like temperature sensing that cannot be replaced with a standard PWM fan. If you run into the same situation as me, just hook up the original fans and you should be good to go.
I don’t know what to make of this yet, but since these fans weren’t the loudest anyway, I am not going to bother with those now.
The general idea is to source ‘silent’ fans and CPU coolers that are able to transport 300W TDP combined. Even though I am not planning on putting the machine under any prolonging stress it will be very rewarding should I at one point decide to do so. Therefore i need to find a way to generate a directional air flow through the case. Here I am aiming for a classic front-pull-rear-push configuration. To make the fans work, i need to figure out what the pinouts for the headers are. Here it will be especially interesting if HPs creative 6-pin solution will pose any threats to my endeavours.
When all that is done, I will have to find a way to get fresh air in through the front. The z820 provides 4x 3.5″ bays and 3x 5.25″ bays for drives and harddisks. Thats all neat, but unfortunately HP did not consider any intake fans at the front and therefore did not leave any space for them. I am pretty positive that I will be getting rid of all 5.25″ drives so that may free up some space to get creative. Apart from getting some space for the fans, I will have to think of a way to get fresh air in through the front without having a gaping hole where once the drives were.
Lastly, I will have to think about cooling VRMs and RAM banks, these can get pretty toasty under prolonging stress.
- source a replacement for the subpar HP AIO cooler (✔)
- source 120mm, 92mm and 80mm fans (✔)
- research pinouts (✔)
- ponder about how to deal with the 6-pin contraption of the rear fans (✔)
- construct a 5.25″ drive cover visually complementing the rest of the z820’s front (✔)
- ponder about how to get air flow to VRM and RAM banks (✔)
- write it all up into something someone might want to read (✘)
After some testing with an expendable fan that I had laying around, it became clear that the plugs HP uses to connect the fans to the fan headers on the fan bracket and the mainboard (with notable exception of the 6-pin exotic) are just different from industry standards because of reasons™. Apparently, HP does not want you to buy cheap, reliable and efficient aftermarket fans that you can just plug in on your own.
Instead, HP wants you to buy cheap, reliable and efficient aftermarket fans where you just have to cut off one of the plugs posts and then plug in:
The different plug layout just seems to exist to disencourage brave people like yourselves from using cheap aftermarket components.
Apart from a small physical modification there are no additional requirements to make the fans work. Just be careful to honor the correct voltage (12V DC) and the maximum current as per the original fans installed.
As to my choice of case fans, I went with the typical 80%-approach. I wanted to have quiet, somewhat-quality budget-friendly fans. After some research, I went with Arctic’s lineup of fans, namely the 80mm, 92mm and 120mm versions of their F-Series. Not only did my research conclude that these were the best choice for me, but also my experience. In earlier projects I already used Arctic PWM fans, and was always pleased with their performance.
All in all I got:
2x 80mm Arctic F8 PWM (rated max. 0.09A)
4x 92mm: Arctic F9 PWM (rated max. 0.09A)
2x 120mm Arctic F12 PWM (rated max. 0.12A)
Regarding new CPU coolers, I decided to go with 2x Arctic Freezer 34 esports Duo. There may be coolers out there with better thermal performance, but I am pretty confident that, in light of better air flow within the case, these will be just fine and capable enough of transporting the heat.
Apart from that, these coolers just fit the geometric constraints of surrounding components such as memory banks, as well as the limits of the case itself. It just so happens that the coolers just barely reach the side panel of the case. Regarding width, they pretty much use up all the available space, without compromising too much serviceability. Both coolers come with two 120mm silent fans, promising to cool these Xeon beasts without becoming too noisy or hot. One pretty neat thing about the ‘Duo’ variant of the cooler is the fact that both fans are connected to the fan header using a single plug. This is achieved by first plugging one of the fans into the receiving female connector of the other. So, while having two fans hooked up, you just need one fan header on the mainboard to supply both.
The 6-pin drama
Yes, well, the elephant in the room. HP decided to create something special, something truly unique to behold. You see, except for the rear case fans, HP just used slightly different plugs for case and CPU fans. So far so good. Not for the rear though. The rear is special. So special in fact, they had to come up with something that befits the occasion. So HP designed this beauty:
So, instead of just having two separate connectors (like with the hdd bay fans) HP opted to go for a combined 6 pin connector. While researching the pinout I came across the z800’s connector pins page. Thankfully, HPs z800 and z820 designs are pretty similar, so I assume that the pinout will be the same for the z820.
Section “Workstation rear system fans, P8” states for the pinout order:
When looking at the cabling, it becomes clear that HP just took two regular PWM fans, soldered each +12V and GND together and stuck them in this non-common plug. That’s it. Thanks, HP.
In order to work non-destructive, I will demonstrate how to create a compatible plug on your own. I did not reuse the original one, simply because I was too lazy to record the assignment of the black, otherwise nondescript cables. You know, just in case to be able to use the original one as backup.
What I did to overcome this obstacle is the following:
I took two regular Arctic F9 PWM fans, and removed the contacts from their plugs. To do that, you just have to look at the plug from the back side. Grab something pointy – pliers or a small knife will do – and push one of the silver contacts while gently tugging the corresponding cable to remove it. Do not apply excessive force!
Once you have done that for both fans, you need to create a new plug for the cables to be plugged into.
To do this, I just reused the fans’ plugs that I removed the cables from. I cut away the superficial posts, gently sanded the sides of the plugs and glued them together using super glue:
You surely will have noticed that I now have a plug with 8 instead of 6 ports, and we all know, more is always better! Jokes aside, I was just way too lazy to cut off 2 unnecessary ports – you can do that, but it won’t hurt to leave them as is – they won’t collide with any modules on the board.
Next, I combined 12V and GND cables of both fans simply by cutting off the contacts on one fan and soldering the castrated cables to the contacts of the other. Do not cut off the PWM and TACH contacts!
All thats left to do is to plug in all the contacts in their respective order. Do not mess this up as this can possibly cause some damage to the fans and/or the mainboard.
This is the finished product:
Well, all that’s left to do now is to replace all original fans with the new ones. Mounting the new CPU coolers was pretty easy, all you have to do is follow the instructions on Arctic’s website.
I will not demonstrate how to do that, since at this point I will just assume that you possess basic computer building knowledge.
A few images of the result:
Having swapped out almost everything cooling-related in this case, we still have to do one last thing. In order for this all to work, we need proper airflow directed through the case. When the design of the workstation was conceived, we all still had one or more optical drives in our computers, preventing an efficient, direct front air intake. Nowadays, we don’t. So why don’t we use the unused space for a solution to the air flow problem?
My solution is as follows: Using FreeCAD, I constructed a 120mm fan mount (*.stl at the end of the article) that clips right into the pins that are supposed to hold 5.25″ drives in place:
Just hook up the 120mm fan to one of the fan headers of the proprietary fan bracket, and you are good to go.
Unfortunately, this solution prevents us from using the drive boxes/front covers, leaving us with an unsightly hole in the front:
Making it pretty again
Having nightmares about this orifice in my beloved z820, I decided to design a visually complementing cover/panel to create a neat front surface, again using FreeCAD (*.stl download at the end of the article). It solves two problems: It closes the case and provides a visually pleasing end to this nightmare of a supposedly simple upgrade and second, it still allows for sufficient air flow through the front. I 3D printed the part with my Creality CR10S printer and sprayed it RAL 9005 matt black using “The Army Painter”.
The cover simply clips into the pockets left and right of the hole:
One more thing…
While testing my new setup I was able to confirm that the machine as a whole has already become significantly quieter, mostly because the new fans are designed more towards low noise level and high efficiency rather than longevity and performance. Since I keep tabs on my hardware on a regular basis, this should not become an issue at all.
Due to the reduced noise level I quickly identified the last noisy culprit in my system: HP decided to disregard any fan control for the rear fans and just powers them 12V, full PWM all the time. Trying to rule out any fault by myself, I hooked up the original 6-pin fan construct with no change in behavior. My assumption therefore is that HP just flat out decided to drive the rear fans with max RPM to create some amount of negative pressure within the case. This makes sense, since it did not have any intake fans directly at the front or somewhere else.
Due to my improved setup, it is just not necessary anymore for this kind of aggressive fan behavior. Since HP does not allow to control individual fans using fan curves in BIOS/EFI or through software, I had to come up with a different solution.
Back in the day, when fan controllers were either non-existent or outrageously expensive, humble pc builders like myself used potentiometers to control the voltage supplied to the fan through variable resistance. Increased resistance means less voltage means slower fans.
This was also encouraged by the fact that PWM fans where either similarly expensive or just not supported by the mainboard.
Feeling 13 again, I went on to calculate the requirements of the potentiometer I was going to use. Since both fans were hooked up in parallel, I had to deal with a rating (maximum heat dissipation) of
P = (0.09A * 2) * 12V = 2.16 VA = 2.16 W
To have some dynamic range in controlling the fan, I decided to step down the voltage to a minimum of 5V. This allows to reduce the fans speed so much as for them to even come to a halt.
This means a reduction of the voltage by 7V. Some quick math:
The basis of all calculations is Ohm’s Law.
Resistance of the fans themselves:
V = I*R ; Ohm's law, therefore R = V/I ; therefore R = 12V/(0.09A*2) R = ~66.6Ω
Since we are targeting a minimum of 5V, we must calculate the target current at 5V first:
V = I*R ; Ohm's law, therefore I = V/R ; therefore I = 5V/66.6Ω I = ~0.075A
Now we need to set the voltage drop in relation to the minimum current to get the required maximum resistance of the potentiometer:
V = I*R => R = V/I => R = 7V/0.075A R = ~93.3Ω => round R = ~100Ω
So now we know, that in order to step down the Voltage by 7V, we need a potentiometer that is capable of dissipating 2.16W of power and that its resistance must be configurable between 0Ω and 100Ω.
A quick search on eBay returned a plethora of results. The one I decided to go with is rated for 4W and 150Ω, so that leaves me with a bit of headroom.
The potentiometer arrived after a few days and I went straight ahead and soldered the whole thing together.
Here is a simple circuit diagram of how everything gets connected:
Please refer to table 2 for the pinout of the mainboards header.
Here you can see the result, already glued into position:
After booting up the machine, i just had to fine tune the resistance using the potentiometers’ knob to something suitable for my use case. Now the rear fans run very quiet but still push lots of air through the rear exhaust. When it gets hotter during summer I will maybe just have to adjust it once to have a little more air moved through the case to compensate for increased ambient temperatures.
One last more thing… the “Press F1 to continue” disaster
Now the last thing that is bugging me is the fact that HP decided to hard code a system halt routine into the firmware that fires when not all fans are present and/or running. I feel this is very pushy, as one can not permanently disable that safety measure. So every time one boots the machine, you have to press F1 to continue the boot process.
To bypass this, I first thought of installing a few additional fans inside the machine. But looking at the already pretty satisfying performance of my custom cooling setup, this seemed unnecessary to me.
So if I am not going to install real fans, why not fake some? To do this, I would have to make the firmware believe that there is a fan running on the unused fan headers.
After some research, I came across NE555 PWM signal generators. These very compact generators create a PWM signal that can be adjusted, too. Many come with voltage converters and therefore can run off of 12V, so you can power them directly off of a fan headers +12V DC source. The PWM signal will then be used to fake running fans by sending it to all the unused TACH pins.
This is the assembly I am using:
If you happen to own one that runs off of 12V, just connect GND and VCC to an unused fan headers corresponding pins and then connect the OUT pin to the TACH pins of all of the unused fan headers. Just make sure that you do a proper soldering job so you don’t short/destroy anything.
And there you have it. My machine is now running much quieter, no gurgling from the liquid cooling pump, the thermal performance is measurably better, no water cooling failure possible anymore and it is easier to service, all in all. My CPUs are idling at ~30°C (86°F) and ~32°C (90°F, ambient room temperature of 20°C, 68° F) at the minimum fan speed setting. Under prolonging synthetic load on all cores the temperatures never rise above 63°C (145°F), thermal throttling never becomes an issue.
UPDATE: Now that it is summer in Germany I once again took a close look at the temperatures. Nothing changed that much. Ambient room temperature has risen to 25° C (77° F), idle temperatures have risen to 33°C and 35°C (91°F and 95°F respectively), synthetic prolonging loads result in temperatures rising to about 68°C – which is still not a concern since Intel specifies a maximum TCASE-Temperature of 72°C and I won’t be running synthetic loads over a long time anyway. And if I do run some memory/CPU-intense workloads, these get load balanced anyway and never stress the CPU that much. So nothing to worry about. CPUs never stop single-core turboing to 3.95 GHz, all-core turbo of 3.6 GHz never gets throttled.
Mission accomplished, I’d say.Citation: me
Tags: HP, z820, workstation, modding, fans, temperature, quiet, pinout, noise, loud
These sources helped me a lot and might be of interest to you, too:
z820 QuickSpecs: https://www8.hp.com/h20195/v2/getpdf.aspx/c04111526.pdf
z820 Maintenance and Service Guide: http://h10032.www1.hp.com/ctg/Manual/c04205252
z800 Connector Pins: https://support.hp.com/us-en/document/c01718096